Anthropic's New AI Helper Sends a Morning Summary and Asks Before It Acts

Anthropic has announced Claude Duet, an AI helper for professionals and small business owners. It is launching as a research preview, which means an early test version for a limited group of people. The Verge reported the announcement on October 21, 2026.

You need to join a waitlist to get access. In the US, you also need a Pro or Max subscription. Approved users can run Duet as a standalone app on iOS and Android, or through Claude on the web. It is built on Claude Opus 5.5.

Duet works without being asked. Each morning, it sends a short brief on the most important items across your connected apps and messages. It contacts you directly only when something needs your attention. To finish tasks, it uses Claude's memory, skills, and connected apps. It also has a virtual machine, a computer program that runs inside another computer, with a web browser it can use.

Before any major step, such as sending a payment or deleting something, Duet asks for your approval. Anthropic has also added a new safety filter, called a classifier, and built-in instructions meant to refuse requests for the most sensitive information. Credit card numbers are given as an example.

Anthropic calls Duet an early version. It may misread a conversation or misjudge what is urgent. It is also weaker at personal errands, such as booking tickets. The company says its urgency judgment is still imperfect.

The Verge describes Duet as a rival to Meta's Muse and OpenAI's Dots AI agents. All three aim at the same need: an assistant that watches several messaging channels and carries out multi-step tasks for you.

Anthropic describes itself as a public benefit corporation, a company formed to secure the benefits of AI and reduce its risks. Its homepage calls it an AI safety and research company that builds systems meant to be reliable, understandable, and controllable. In March, CIO Dive reported that Anthropic had let Claude use a computer from a mobile device, in a research preview for Pro and Max users. Duet continues that path into a role that runs over time across several apps.

In my view, the approval step and the safety filter are the parts to watch most closely. Older assistants mostly gave wrong answers that you could catch before anything happened. An agent that clicks through a browser and can reach payment pages can turn a mistake into an action, and that is harder to undo.

The broader context here is that AI safety is shifting from what a system says to what it does for you.

The upside, as I see it, is easy to picture. A small business owner who starts the day sorting email and calendars could get time back, if the morning brief reliably shows only the items that need a decision. Whether it does will depend on how well those urgency judgments hold up beyond a controlled test, which is exactly what a research preview is for.